基于优化字典设计的MOD字典学习算法  

Design of MOD Dictionary Learning Algorithm Based on Optimized Dictionary

作　　者：周航 董西伟[1,2] 荆晓远[1] 

机构地区：[1]南京邮电大学自动化学院,江苏南京210023 [2]九江学院信息科学与技术学院,江西九江332005

出　　处：《计算机技术与发展》2018年第1期56-59,共4页Computer Technology and Development

基　　金：国家自然科学基金资助项目(61272273)

摘　　要：在模式识别研究领域,有关人脸识别的研究一直备受关注,并且已经成功地应用于诸多社会公共安全防护领域。近年来,随着压缩感知理论的发展,稀疏表示因其出色的分类性能以及对噪声因素的鲁棒性而受到众多研究者的关注,并且被成功地应用于人脸识别当中。基于稀疏表示的分类算法的性能优劣与学习到的字典息息相关,因此字典的优化设计非常值得深入研究。文中在经典的MOD算法中加入聚类算法,提出一种增强型MOD字典学习算法(E-MOD)。该算法在字典学习阶段使用聚类算法来优化字典的设计,去除冗余的字典原子数,得到性能优秀的字典;接着为了使学习到的字典具有判别性能,进一步使用MOD算法继续学习,最终得到分类效果更佳的字典。在AR和CAS-PEAL人脸数据库上的对比实验有效地验证了E-MOD算法的性能。In pattern recognition, research on face recognition has attracted much attention, with successful application to many areas of so- cial public security protection. Recently, with the rising of compressive sensing, sparse representation has received extensive attention be- cause of its excellent classification and robustness to noise, which has been successfully used in face recognition. The performance of sparse representation based classification is closely to the learned dictionary, so optimized design of dictionary is a very worthwhile study project. Adding clustering algorithm into the classical MOD algorithm, an Enhanced MOD （E-MOD） algorithm is proposed in this pa- per. The clustering algorithm is used to optimize the design of dictionary in the dictionary learning phase, removing the redundant diction- ary atoms, getting the better performance dictionary. Then with MOD algorithm to further continue learning in order to make the dictiona- ry are discriminative ,it can make the dictionary has superior classification effect. The comparison experiments on the AR and CAS-PEAL face databases have verified the effectiveness of the proposed algorithm.

关 键 词：稀疏表示 字典学习 MOD算法 竞争聚集 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]